Re: orbit-small.c patch

[Michael Meeks <michael ximian com>]

Hi Frank,

On Tue, 2003-06-10 at 22:54, Frank Rehberger wrote:
> If skeleton-implementation (badcall-sleimpl.c)  raises exception the  
> memory allocated for arguments has not been CORBA_free'd on return.  
>  This patch removes two guards, so the memory is free'd in any case.

	Sounds dodgy to me.

> As this patch assumes that servant-operations return valid values  and 
> out-arguments it might have impact on existing servant-implementations, 
> that handle return values lazy in case of exceptions.

	I believe the spec. says that return values from skels are undefined on
exception ( as are out arguments ).

	Thus - freeing undefined values will ~result in yet more undefined (ie.
acutely bad) behavior.

	So - the trick is to cleanup properly before returning an exception in
your skel impl.

	HTH,

		Michael.

-- 
 michael@ximian.com  <><, Pseudo Engineer, itinerant idiot